Skip to content


ai  101  pytorch  classification  nvidia  cuda  install  tensorrt  yolo  ardupilot  None  ros2  dds  micro ros  xrce  sitl  plugin  SITL  debug  rangefinder  pymavlink  mavros  gazebo  distance sensor  system_time  timesync  cmake  gtest  ctest  cpp  c++  format  fmt  multithreading  spdlog  camera  coordinate system  orb  matching  opencv  build  transformation  computer vision  homography  optical flow  of  trackers  cv  cyclonedds  eprosima  fastdds  simulation  config  ignition  bridge  sdf  tips  ign-transport  sensors  lidar  aptly  apt  encryption  pgp  docker  git  bundle  github  hooks  pre-commit  lxd  container  lxc  x11  profile  vscode  marpit  presentation  marp  markdown  mermaid  video  ffmpeg  gstreamer  cheat-sheet  sdp  v4l2loopback  gi  snippets  cheat Sheet  python  asyncio  future  click  cli  numpy  project  template  black  isort  docs  project document  docstrings  flake8  linter  git-hook  mypy  unittest  pytest  pylint  mock  iterator  generator  logging  tuple  namedtuple  typing  annotation  pyzmq  zmq  msgpack  action  namespace  remap  control2  ros2_control  gdb  qos  tag  plugins  msg  node  zero-copy  shm  tutorial  algorithm  calibration  diff  pid  dev  colcon  colcon_cd  rpi  arm  qemu  settings  behavior  plot  visualization  debugging  diagnostic  diagnostics  tutorials  gst  math  apm  rat_runtime_monitor  web  rosbridge  vue  binding  discovery  gazebo-classic  launch  spawn  cook  gps  imu  ray  gazebo_ros_ray_sensor  ultrsonic  range  ultrasonic  gazebo classic  wrench  effort  odom  ign  gz  xacro  ros_ign  diff_drive  odometry  joint_state  argument  OpaqueFunction  DeclareLaunchArgument  LaunchConfiguration  tmux  nav  slam  test  rclpy  executor  MultiThreadedExecutor  SingleThreadedExecutor  param  dynamic-reconfigure  service  client  setup.py  package.xml  parameter  parameters  custom  msgs  executers  pub  sub  rqt  rviz  rviz2  pose  marker  tf2  deb  package  setup  local_setup  rosdep  package manager  project settings  vcstool  cross-compiler  nano  texture  tmuxp  rootfs  embedded  zah  linux  rm  ubuntu  ip  ss  network  netstat  snap  deploy  ssh  systemd  mkdocs  extensions  socat  networking  serial  udp  tc  mtu  select  px4  robotics  kalman_filter  kalman  filter  control  todo  vscode-ext  json  yaml  schema  yocto  poky  world  gazebo_ros2_control  position_controller  effort_controller  velocity_controller  urdf  gazebo_ros_force  gazebo_ros_joint_state_publisher  robot_state_publisher  joint_state_publisher  projects  vrx  buoyancy 

SiamMask

SiamMask

#ai   #trackers   #cv  

Table of Content

SiamMask#

SiamMask used to tracking and segment objects from videos in each frame, initializing a single bounding box and outputing binary segmentation mask and rotated objects boxes

SiamMask needs to be initialized with a single bounding box so it can track the desired object. However, this also means that multiple object tracking (MOT) is not viable with SiamMask

alias#

  • VOT: Visual Object Tracking
  • MOT: Multiple Object Tracking
  • VOS: Video Object Segmentation
  • SOT: Signal Object Tracking

Demo#

  • Clone project
  • Setup environment
  • Download model
  • Run
clone
git clone https://github.com/augmentedstartups/SiamMask.git

!!! Note title=”requirements.txt” Running different python package version from github settings

Cython==0.29.28
colorama==0.4.4
numpy==1.21.5
requests==2.22.0
fire==0.4.0
torch==1.5.1+cu101
matplotlib==3.5.1
numba==0.55.1
scipy==1.7.3
h5py==3.6.0
pandas==1.4.0
tqdm==4.64.0
tensorboardX==2.5
torchvision==0.6.1+cu101

**build from source**
opencv 4.5.4

cd SiamMask
bash make.sh
setp env
# Add SiamMask root project and demo folder siammask_sharp folder to PYTHONPATH
cd SiamMask
export PYTHONPATH=`pwd`:$PYTHONPATH

cd experiments/siammask_sharp
export PYTHONPATH=`pwd`:$PYTHONPATH
download models
cd SiamMask/experiments/siammask_sharp
wget http://www.robots.ox.ac.uk/~qwang/SiamMask_VOT.pth
wget http://www.robots.ox.ac.uk/~qwang/SiamMask_DAVIS.pth

bash title="run python ../../tools/demo.py --resume SiamMask_DAVIS.pth --config config_davis.json

Note

DAVIS_2016: Video object segmentation dataset
DAVIS16 is a dataset for video object segmentation which consists of 50 videos in total (30 videos for training and 20 for testing). Per-frame pixel-wise annotations are offered.
dataset

Reference#

To check#

SenseTime Research platform for single object tracking, implementing algorithms like SiamRPN and SiamMask - pysot